I used to think that "em" units were the bee's knees. The cat's meow. The way it should be done.
Then I encountered weird bugs and design glitches and spent hours trying to figure out how to fix them. IE 5.5 was a real bear to whip into submission. IE 6.0 has some trouble with em units too.
Then I saw how Opera handled text zooming by zooming the whole page, rather than just increasing the text size. IE7 followed suit. And now Firefox 3 will support it too.
So I came to a rather obvious conclusion. The simple fact of the matter is that the overwhelming vast majority of websites do not implement designs in em units and never will. At that only a fraction of the percentage of people implementing bonafide, best practice oriented, standards-driven designs will bother with em units because of the design complications it introduces.
On the surface it's a cool idea, and makes you feel good because you're promoting an accessible design. But it doesn't work in the real world, and is a feature of CSS that is doomed to failure because it puts the burden on the website designer/developer rather than the browser, and it's a burden you can completely ignore, so most people do.
When you give it any thought you realize that page zooming is the correct way to approach the problem, because it works for every website, even the bloated, non-standard crap infesting this web's hallowed halls, rather than only the handful that take text resizing into account.
Visual disabilities are pretty common. There are a lot of people who have trouble reading text... however... the "em" battle is a battle that cannot be won... and page zooming is a superior solution to the problem that helps people with visual disabilities surf the whole web without hindering a single site. If I know someone with visual disabilities that need to be able to zoom text, I'd recommend Firefox 3, IE7, or the latest Opera.
Page zooming should be standard. em units (and similar units) should at the very least be abandoned for this purpose if not deprecated all-together.